AI-Generated Tigers Shed Chinese Influence... 9,000 Korean Folk Painting Data Points Compiled

National Heritage Promotion Agency Unveils Results of "Traditional Folk Painting Data" Project
Correcting AI's "Style Distortion"... Full Release on "AI Hub" Planned for First Half of the Year

Fake painting created by AI

Until now, generative artificial intelligence (AI) has shown its limitations by producing Chinese-style paintings or images with ambiguous origins when asked to create “Korean folk paintings.” To address this technical distortion and teach AI the unique aesthetic of Korea, a large-scale dataset has been developed.

The National Heritage Promotion Agency held a performance report session for the “Korean Traditional Folk Painting Data Production Project” at Hotel Crescendo in Seoul on January 20. This achievement is part of the “2025 Hyper-scale AI Ecosystem Expansion Project” led by the Ministry of Science and ICT and the National Information Society Agency (NIA). HCI Plus, the Gahoe Folk Painting Museum, and Jeju National University participated in a consortium and conducted this project over the past year.

The key objective is to enable AI to accurately recognize and reproduce the distinctive painting styles and motifs of Korean folk art. To this end, the research team secured a total of 9,119 high-quality images, including 3,779 subject-specific folk paintings such as flower-and-bird paintings, landscape paintings, and chaekgado, along with 5,340 detailed depiction images. To further enhance learning precision, they combined these images with 77,388 Korean-English multimodal caption data points, allowing AI to better understand context. An official from the agency stated, “We ensured the reliability of the data by verifying the artists’ eras and motif systems based on museum collections.”

During the report session, an AI model trained on the newly built dataset was also demonstrated. When a text prompt was entered, the AI generated images in real time that captured the unique textures and colors of traditional folk paintings. Demonstrations included transforming the tiger character (Duffy) from the animated series “K-Pop Demon Hunters” into a folk painting style, confirming the potential for traditional cultural heritage to be fused with modern content.

Tiger character created with AI

The agency also released prototypes of four cultural products created using the dataset, including art lamps featuring tiger-and-magpie paintings and flower-and-bird paintings. Various projects are planned to expand the use of this dataset beyond research, into commercial fields such as industrial design, product development, and media art.

The constructed folk painting dataset will be fully opened through “AI Hub” in the first half of this year. Shim Jeongtaek, head of the data team at the National Heritage Promotion Agency, stated, “This project not only confirmed the potential for integrating our folk paintings with AI, but also reaffirmed the importance of high-quality training data. We will continue to expand the dataset to enhance the value and utilization of our cultural heritage.”